抄録
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (B), Fund Type: -, Overall Grant Amount: - (direct: 11600000, indirect: -)
In this research, we have developed a novel set of methodology that makes autonomous robots act in the real-world for long time. For technology exchange, also, we have participated in the robot soccer world cup (RoboCup), where researchers have make robots work in the real-world successfully.
The methods developed are as follows: self-localization that is robust against unexpected conditions, a decision making method that cover any condition in a task, and another decision making method that deals with uncertainty of recognition.
Our novel self-localization method, which is a version of a particle filter, utilizes resetting methods. A resetting method is used when a particle filter loses the pose of a robot. We enhance the robustness of a particle filter for self-localization by combining two major resetting methods, which have antipodal characteristics.
As a planning and decision making method, we have proposed a compression method for decision making data, which is huge for install on memory of robots. Vector quantization is used for the compression.
The particle filter and the decision making rule is combined so that robots can deal with the problem of uncertainty in decision making. In this method, an expected value of appropriateness is calculated for each action toward the probability distribution represented by the particle filter. The most appropriate action is then chosen from the values. In experiments, robots can switch their behavior when the level of uncertainty is changed. For example, a robot keeps away from a wall of an environment when the distance between the robot and the wall is uncertain.